Each dataset is splitted by trainset, devset and testset.

Please read them with pytorch.